CS780: Scalable Graphics/Geometric Algorithms- Summary

Sung-Eui Yoon(윤성의)

Course URL:http://jupiter.kaist.ac.kr/~sungeui/SGA/

2

In a Nutshell●We have studied:

● Various techniques to design scalable graphics algorithms that can handle massive models

3

Geometric Data Explosion● Massive geometric data

● Due to advances of modeling, simulation, and data capture techniques

44

Large-scale Applications● Entertainment (games/movies)

● Second Life

55

Large-scale Applications● Computer-aided design (CAD) / virtual

prototyping

Ray Tracing Boeing 777,470 million triangles

Excerpted from SIGGRAPH course note on massive model rendering

66

Large-scale Applications● Geographic information system (GIS)

● Google earth

77

Large-scale Applications● Robotics

● Motion planning

Excerpted fromMotion Planning in Virtual Environments and Games, Mark Overmars

8

Possible Solutions?●Hardware improvement will address the

data avalanche?● Moore’s law: the number of transistor is

roughly double every 18 months

9

Current Architecture Trends

1

10

100

1000

Diskaccessspeed

RAMaccessspeed

CPUspeed

GPUspeed

Accumulated growth rates

during 1993 – 2004(log scale)

during 99 - 04

1.5X

20X46X

130X

Courtesy: Anselmo Lastra, http://www.hcibook.com/e3/

online/moores-law/

Data access time becomes the major computational bottleneck!

10

Current Architecture Trends: Many-cores● Employs multi-cores to keep Moore’s law

● 80 core system in Intel● Presents numerous research challenges

● Streaming processors (GPUs) with super Moore’s law● Multi stages in parallel

Data access time is getting relatively bigger!

11

Data Growth● An observation

● If we got better performance, we attempt to produce bigger data to derive more useful information and handle such bigger data

● Amount of data is doubling every 18 ~ 24 months● “How Much Information,” 2003. Lyman, Peter

and Hal R. Varian., www.sims.berkeley.edu/how-much-info-2003

12

Ubiquitous Computing●Uses different computational devices

● Have relative small main memory and L1/L2 caches

● Pose problems even with small models

Google Earth: browsing 3D world

13

Our Focus in the Course● Technologies for scalable graphics

applications● Multi-resolution methods● Cache-coherent algorithms● Culling techniques● Selective construction methods● (Data compression, parallel computations, etc)

●Graphics applications● Rasterization and ray tracing for rendering● Collision detection

Use simplification given an error bound

Reduce the model complexity!

Multi-Resolution or Levels-of-Detail (LODs)

Culling Techniques● Visibility culling

ViewerVisible

trianglesInvisibletriangles

Cull away

Reduce the model complexity!

Data organization to reduce expensive data access time!

Cache-Coherent Layouts

va

vb vdvc

va vb vd vc

One dimensional layout

17

Selective Restructuring

BVH

RestructureRefit

18

Uncovered Topics●Data compression

● Data size is still too large● Parallel computations

● Many cores are available● Cache-coherent algorithms

● We only covered coherent data layouts

● Various other graphics/geometric related applications

19

Administrations●Give you feedbacks on your reports by early

next week● Grade will be given after that

● Course evaluations● Let’s make this course better for coming

students!!!

20

Thanks!